Skip to content

Plainsong 0.9.0 beta 4

Pre-release
Pre-release

Choose a tag to compare

View all tags
@JonathanRReed JonathanRReed released this 05 Sep 01:15
v0.9.0-beta.4
5c0ca09
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Plainsong 0.9.0 Beta 4 is an internal Apple Silicon beta.

What changed:

  • Consolidated the 75-PR audit backlog: 73 merged and 2 closed as superseded.
  • Hardened process execution, temporary audio handling, dependency versions, and secret scanning.
  • Added Parakeet V2 as an experimental local dictation option while keeping Parakeet V3 as the recommended route.
  • Added the requested local Ollama catalog: GPT-OSS 20B, DeepSeek R1 Distill 8B, Ministral 3 8B, Llama 3 8B, Mistral 7B v0.2, Llama 3.2 3B, and Phi-2 3B.
  • Added or retained local and BYOK transcription routes for Cohere, Whisper, ElevenLabs Scribe V2, and Deepgram Nova 3, Nova 2, and Nova Medical.
  • Kept on-device processing as the default. Cloud processing remains explicit BYOK opt-in.
  • Fixed a macOS Apple Speech helper pipe deadlock found during release qualification.

Release evidence:

  • Exact source snapshot passed all 10 source gates.
  • 1,897 frontend tests and 1,731 Rust tests passed.
  • JavaScript and Rust dependency audits reported no known vulnerabilities.
  • Signed and notarized with Developer ID for Apple Silicon. Gatekeeper and stapling validation passed.
  • Packaged first-run, dictation hotkey, mic meeting capture, and combined mic plus system-audio capture passed.
  • Voz was evaluated and is not included because its available terms require attribution and non-disableable usage telemetry.

Known limits:

  • This build is not qualified for Intel Macs.
  • Parakeet V2 and the large Ollama downloads are cataloged but have not all been exercised end to end on this candidate.
  • Competitive superiority over Granola, Superwhisper, Raycast Dictation, or Wispr Flow is not claimed without same-hardware benchmark evidence.
  • The Half-Bounce launch target is not met yet. The recorded candidate run occurred under extreme host load and failed the 1.5-second gate.
  • Hosted GitHub Actions could not allocate runners because of the account billing or spending-limit state. Local release gates and Macroscope review passed.
Assets 6
Loading